ORBIT-FMIB: Tracking Order-Resolved Epistatic Information Through ESM-2

2026-09-30

AI Summary

This study addresses the limited interpretability of high-order epistatic interaction structures in protein foundation models by investigating whether ESM-2 retains high-order epistatic information. Methodologically, it introduces the first high-order interaction diagnostic protocol tailored for protein foundation models and proposes the ORBIT-FMIB framework, which integrates Walsh interaction decomposition, subset-conditioned neural dependence estimation, and frozen representation analysis for systematic evaluation. The findings reveal the critical impact of reproducibility on the robustness of conclusions and demonstrate that current evidence is insufficient to assert that ESM-2 selectively discards high-order information. Consequently, this work emphasizes that future validation efforts must incorporate adequately powered replication checks to ensure statistical rigor.

Abstract

Protein foundation models support mutation-effect and structural prediction, but predictive performance alone does not reveal which forms of biological interaction information remain accessible through model depth. We ask whether ESM-2 retains higher-order epistatic information as strongly as first- and second-order information across its representation hierarchy, introducing ORBIT-FMIB, a diagnostic framework combining Walsh-based interaction decomposition with subset-conditioned neural dependence estimation. The method is validated on synthetic landscapes with known interaction structure before being applied to the dense four-site GB1 fitness landscape using frozen ESM-2 representations. An initial production run suggested ESM-2 retains higher-order epistatic information less well than lower-order information ($Δ_{\mathrm{HO-LO}}=-0.107$). An independent replication of the complete measurement grid, under matched GPU hardware and identical critic seeds, substantially reduced this contrast ($Δ_{\mathrm{HO-LO}}=-0.017$), and its sign was unstable across otherwise-defensible evaluation-pairing choices applied to the same trained critics ($-0.011$ to $+0.015$). We therefore do not currently have robust evidence that ESM-2 selectively loses higher-order epistatic information, nor that retention is equal across orders; the directional question remains open. The measurement protocol itself, including its documented removal of a positional-subset shortcut in pooled critics, remains validated and is unaffected by this finding. ORBIT-FMIB is offered as a diagnostic framework for probing interaction structure in protein foundation models; this study's own replication result illustrates why such probing requires adequately-powered reproducibility checks before its output is treated as a biological finding.

Problem

• protein foundation models
• higher-order epistasis
• ESM-2
• interaction information
• representation hierarchy

Innovation

• ORBIT-FMIB
• Epistatic information
• Walsh-based interaction decomposition
• Neural dependence estimation
• Protein foundation models